1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
|
profnet
=======
profnet_* binaries are neural network implementations in Fortran.
Due to the original design of the code, a specific binary is compiled
for each particular network architecture, changing certain constants
in the source code. Therefore, there is a binary for every network
architecture used. Note: certain array structures are intentionally
indexed out of bounds in some of the binaries!
Q: Why so many binary packages?
-------------------------------
There are a handful of prediction methods built around each of the binary
profnet packages. Each depends on the matching profnet binary package.
Each prediction method requires a different neural network architecture
and therefore - due to the design of the code - a different binary.
Binaries are compiled with constants set to the architecture of
the network and are therefore not reusable for other architectures.
To further develop the code beyond regular maintenance for compiler and
architecture updates is not planned since a complete reimplementation
of these networks with a neural network library is already underway.
Publications of predictors that use these neural networks
---------------------------------------------------------
Note: this list contains references only to the secondary structure and
accessibility predictor (profnet-prof and profphd-net).
References for the other methods are provided in the man pages of the
respective methods' predictor commands, e.g. profbval(1).
* Rost, B. and Sander, C. (1994a). Combining evolutionary information and neural networks to predict protein secondary structure. Proteins, 19(1), 55-72.
* Rost, B. and Sander, C. (1994b). Conservation and prediction of solvent accessibility in protein families. Proteins, 20(3), 216-26.
* Rost, B., Casadio, R., Fariselli, P., and Sander, C. (1995). Transmembrane helices predicted at 95 Protein Sci, 4(3), 521-33.
-- Laszlo Kajan <lkajan@rostlab.org> Tue, 14 Jun 2011 18:50:52 +0200
|
